baskduf/FableCodex
🗿 FableCodex is a Codex-style coding agent workflow that plans like Fable.
What is FableCodex?
FableCodex is a Codex plugin that adds a disciplined, evidence‑based workflow to any task you ask Codex (the code‑generation model) to perform. Instead of letting the model edit files unchecked, the plugin makes Codex:
- Classify the request before acting.
- Inspect the current workspace and any referenced sources.
- Use the appropriate Codex‑native tools (e.g., file‑edit, test runner) rather than relying on memory.
- Track a multi‑step goal ledger, recording evidence at each checkpoint.
- Collect review findings and keep them open until they are explicitly resolved.
- Verify the result with tests, linters, type‑checks, screenshots, command output, etc.
- Report what changed, what was verified, and what risk remains.
The plugin does not modify the underlying model, its weights, or its safety systems – it only adds procedural guidance and local bookkeeping.
Who might want to use it?
| Good fit | Not a good fit |
|---|---|
| Multi‑step implementations, refactorings, or migrations | One‑line fixes or tiny edits |
| Debugging hard‑to‑reproduce failures | Simple brainstorming or idea generation |
| CI failures, release work, security‑sensitive changes | Tasks where the extra overhead outweighs the benefit |
| Reviews where findings must block completion | Quick “look‑and‑say‑done” answers |
| Teams that already have access to a Codex model and want stricter process | Users without a Codex integration |
How does a user control the workflow?
Control is done entirely through the prompt you send to Codex, using the @codex-fable5 tag. Examples:
- Strict mode – full ledger, findings gate, and verification before finishing:
@codex-fable5 Run this strictly. Use a goal ledger, record any review findings, and do not finish until tests and findings gate pass. - Analysis only – no file edits, just a report of findings:
@codex-fable5 Analyze only. Do not edit files. Give findings with file and line references. - Implementation with limits – edit but don’t push or delete branches:
@codex-fable5 Implement the fix. Do not commit, push, or delete branches. Run unit tests and report any residual risk. - Debugging – reproduce, hypothesise, gather disconfirming evidence, then fix:
@codex-fable5 Debug this failure. Reproduce it first, keep multiple hypotheses, gather disconfirming evidence, then fix and verify.
Core concepts
Goal Ledger
- Stored locally in
.codex-fable5/goals.json. - Created with
codex-fable5 goals create …specifying a brief description and a list of sub‑goals. - Each sub‑goal is marked complete with
codex-fable5 goals checkpoint …, supplying evidence (e.g., “read file X, updated tests”) and optionally a verification command and its output.
Findings Gate
- Review issues are recorded in
.codex-fable5/findings.jsonviacodex-fable5 findings add …. - A finding stays open until it is resolved with
codex-fable5 findings resolve …, which also requires verification evidence. codex-fable5 findings gateaborts final completion if any high‑severity or blocked findings remain.
Quick start (from the README)
# Install the latest stable release
codex plugin marketplace add baskduf/FableCodex --ref v0.5.1
codex plugin add codex-fable5@fablecodex
# Restart Codex, then invoke the skill in your prompt
@codex-fable5 Use this skill to implement the change.
Create a goal ledger if the work has multiple steps.
Track findings before final completion.
Run the project tests before saying it is done.
A lighter version skips the ledger:
@codex-fable5 Review this quickly.
Do not create a goal ledger. Check the key evidence and report only actionable findings.
Command reference (summary)
| Command | What it does |
|---|---|
codex-fable5 status |
Shows current goals and findings |
codex-fable5 goals create … |
Starts a new multi‑step ledger |
codex-fable5 goals next |
Shows the next pending goal |
codex-fable5 goals checkpoint … |
Marks a goal as complete/failed with evidence |
codex-fable5 findings add … |
Records a review finding |
codex-fable5 findings resolve … |
Closes a finding with verification evidence |
codex-fable5 findings gate |
Fails if any open/blocked findings remain |
codex-fable5 update |
Pulls the latest stable tag of the plugin |
Additional features
- Coverage accounting – if you have the original
CLAUDE‑FABLE‑5.mdsource, a helper script can verify that every source heading has a corresponding decision (implemented, adapted, unsupported, or N/A). - Optional provider bridge – documentation is provided for routing requests through Anthropic or an OpenAI‑compatible gateway (e.g., LiteLLM). The plugin itself does not ship any model credentials.
- Local state – all ledgers and event logs live under
.codex-fable5/and are meant to be transient; commit only if you deliberately want a permanent transcript.
License & community
- Licensed under AGPL‑3.0‑or‑later.
- Recognised in community skill lists such as
antigravity-awesome-skillsandawesome-agent-skills. - Contributions are welcomed; a contributors badge is displayed in the README.
TL;DR
FableCodex is a process‑oriented wrapper for the Codex code‑generation model. By making the model follow a checklist‑style workflow—goal ledgers, evidence checkpoints, findings gates, and automated verification—it helps teams avoid missed steps in larger, safety‑critical code changes while staying completely within the Codex ecosystem.
Related
- Dispatch
- Project
- Project
- Dispatch
- Project